diff --git a/include/haproxy/thread.h b/include/haproxy/thread.h
index ad0ae1aa7..c030c5e0f 100644
--- a/include/haproxy/thread.h
+++ b/include/haproxy/thread.h
@@ -360,7 +360,15 @@ static inline unsigned long thread_isolated()
 	} while (0)
 
 #define _lock_wait(_LK_, lbl, expr) do {				\
+		uint64_t lock_start = 0;				\
+		extern uint64_t now_mono_time(void);			\
+		if (_LK_ != _LK_UN) {					\
+			if (unlikely(th_ctx->flags & TH_FL_TASK_PROFILING)) \
+				lock_start = now_mono_time();		\
+		}							\
 		(void)(expr);						\
+		if (_LK_ != _LK_UN && unlikely(lock_start))		\
+			th_ctx->lock_wait_total += now_mono_time() - lock_start; \
 		if (lbl != OTHER_LOCK)					\
 			_lock_wait_common(_LK_, lbl);			\
 	} while (0)
diff --git a/include/haproxy/tinfo-t.h b/include/haproxy/tinfo-t.h
index 9e379ae5b..78084c497 100644
--- a/include/haproxy/tinfo-t.h
+++ b/include/haproxy/tinfo-t.h
@@ -161,13 +161,16 @@ struct thread_ctx {
 
 	uint32_t sched_wake_date;           /* current task/tasklet's wake date in 32-bit ns or 0 if not supported */
 	uint64_t sched_call_date;           /* current task/tasklet's call date in ns */
+	uint64_t lock_wait_total;           /* total time in ns spent waiting for a lock (task prof) */
 
 	uint64_t prev_mono_time;            /* previous system wide monotonic time (leaving poll) */
 	uint64_t curr_mono_time;            /* latest system wide monotonic time (leaving poll) */
 
 	ulong lock_history;                 /* history of used locks, see thread.h for more details */
 
-	// third cache line here on 64 bits: accessed mostly using atomic ops
+	/* around 56 unused bytes here */
+
+	// fourth cache line here on 64 bits: accessed mostly using atomic ops
 	ALWAYS_ALIGN(64);
 	struct mt_list shared_tasklet_list; /* Tasklet to be run, woken up by other threads */
 	unsigned int rqueue_ticks;          /* Insertion counter for the run queue */
diff --git a/src/task.c b/src/task.c
index 72be355f0..fdd2775ba 100644
--- a/src/task.c
+++ b/src/task.c
@@ -569,6 +569,7 @@ unsigned int run_tasks_from_lists(unsigned int budgets[])
 		process = t->process;
 		t->calls++;
 
+		th_ctx->lock_wait_total = 0;
 		th_ctx->sched_wake_date = t->wake_date;
 		if (th_ctx->sched_wake_date || (t->state & TASK_F_WANTS_TIME)) {
 			/* take the most accurate clock we have, either
@@ -678,8 +679,11 @@ unsigned int run_tasks_from_lists(unsigned int budgets[])
 		__ha_barrier_store();
 
 		/* stats are only registered for non-zero wake dates */
-		if (unlikely(th_ctx->sched_wake_date))
+		if (unlikely(th_ctx->sched_wake_date)) {
 			HA_ATOMIC_ADD(&profile_entry->cpu_time, (uint32_t)(now_mono_time() - th_ctx->sched_call_date));
+			if (th_ctx->lock_wait_total)
+				HA_ATOMIC_ADD(&profile_entry->lkw_time, th_ctx->lock_wait_total);
+		}
 	}
 	th_ctx->current_queue = -1;
 	th_ctx->sched_wake_date = TICK_ETERNITY;